1

Dynamic Trust Enforcing Pricing Scheme forSensors-as-a-Service in Sensor-Cloud

InfrastructureAishwariya Chakraborty†, Student Member, IEEE , Ayan Mondal†, Student Member, IEEE ,

Arijit Roy‡, Student Member, IEEE , and Sudip Misra†, Senior Member, IEEE

Abstract—In this paper, the problem of provisioning high quality of Sensors-as-a-Service (Se-aaS) in the presence ofcompetitive sensor-owners, i.e., oligopolistic market, and heterogeneous sensor nodes in service-oriented sensor-cloud isstudied. Oligopolistic sensor-owners adopt unfair means to degrade the quality of service provided by other sensor-ownersin the sensor-cloud market. In order to address this problem, a dynamic pricing scheme, named DETER, is proposed in thiswork to enforce trust among the sensor-owners for maintaining the quality of Se-aaS provided by the Sensor-Cloud ServiceProvider (SCSP). Each sensor node calculates distributed trust opinion for other nodes, while the SCSP calculates centralizedtrust opinion for each sensor-owner. A Single-Leader-Multiple-Follower Stackelberg Game is formulated in which the SCSP actsas the leader and decides price to be paid to each sensor-owner, while ensuring maximum profit. On the other hand, the sensor-owners act as the followers and decide their strategies for earning maximum profit. Thereby, using DETER, SCSP enforces hightrust among the sensor-owners. Additionally, using DETER, energy consumption of sensor nodes in sensor-cloud decreases by4.69-11.56%, and network overhead decreases by 52.6-56.53%. The trade-off between price earned by the sensor-owners andprofit of the SCSP in service-oriented sensor-cloud is also maintained using DETER.

Index Terms—Sensor-Cloud, Trust Enforcing, Dynamic Pricing, Oligopoly, Bi-Level Trust, Game Theory.

F

1 INTRODUCTION

FOR the development of a wireless sensor network(WSN)-based application, a user is generally re-

quired to purchase, deploy and maintain his/her ownapplication-specific hardware resources. So, in orderto relieve the users of WSN-based applications fromthe associated implementation complexities and fi-nancial expenditures, the sensor-cloud architecture wasconceptualized [1]. Sensor-cloud basically follows aService-Oriented Architecture (SOA). In sensor-cloud,the responsibilities of a user are shifted to a cen-tralized Sensor-Cloud Service Provider (SCSP), whichobtains WSNs on rental basis from multiple sensor-owners and provides these resources to the users inthe form of chargeable units of services. The conceptof resource virtualization of cloud computing is ex-tended into WSNs to allow sharing of physical sensornodes between multiple end-users, thereby allowingthe provisioning of Sensors-as-a-Service (Se-aaS). Bytransferring majority of the functionalities from thesensor nodes to the cloud, sensor-cloud reduces thecomputational burden as well as the resource con-

• † Aishwariya Chakraborty, Ayan Mondal, and Sudip Misra arewith the Department of Computer Science and Engineering,Indian Institute of Technology Kharagpur, India (Email:aishwariyachakraborty@gmail.com; ayanmondal@sit.iitkgp.ernet.in;smisra@sit.iitkgp.ernet.in).

• ‡ Arijit Roy is with the Advanced Technology Development Cen-tre, Indian Institute of Technology Kharagpur, India (Email: ari-jitr@iitkgp.ac.in

sumption of each node, while increasing the over-all network lifetime. Additionally, sensor-cloud alsoprovides a platform for sensor-owners to increasethe utilization of their hardware resources and earnprofits by leveraging the same.

1.1 Motivation

Sensor-cloud provisions Se-aaS to the end-users andinvolves a flow of revenue from the end-users tothe SCSP. In the competitive Se-aaS market, an SCSPneeds to ensure proper pricing and quality of thedelivered service. Chatterjee et al. [2] envisioned Se-aaS as a combination of both hardware and infrastruc-ture services — heterogeneous SOA, unlike traditionalcloud services. Hence, the existing pricing schemesdesigned for WSNs [3], [4] and cloud services [5], [6]are not suitable for sensor-cloud. On the other hand,for providing high quality-of-service (QoS) in sensor-cloud, it is indispensable for an SCSP to maintainaccuracy or correctness of the data requested by anend-user. However, this issue is not addressed in theexisting literature. In sensor-cloud, a fraction of therevenue earned by the SCSP is distributed amongthe sensor-owners for usage of their sensor nodesfor sensing and relaying of sensed data based oncentralized decision taken by the SCSP. This gives riseto an oligopolistic market scenario among the sensor-owners in the SOA of sensor-cloud. In sensor-cloud,the sensor-owner of the source node has to rely on

2

the intermediate nodes, which are selected in a dis-tributed manner, belonging to other sensor-owners totransmit the sensed data to the base-station (BS). Inthe process, these sensor-owners gain complete con-trol over the sensed data. In such a situation, a selfishsensor-owner, who is driven solely by the urge tosatisfy his/her self-interests, may adopt unfair meansto reduce QoS. Thereby, the reputation of the honestsensor owner reduces, which in turn reduces the se-lection preference of the nodes belonging to him/her.This may eventually give rise to a monopoly situation.Additionally, the net profit incurred by the SCSPreduces. In the existing literature, few works exist onsensor-cloud, which include service-oriented pricingschemes, viz., [2]. However, none of these works con-sidered the possibility of the aforementioned event.Hence, it is important to design a scheme, which issuitable for the sensor-cloud environment, in order toprevent such kind of monopoly situation from arisingin oligopolistic market scenario of sensor-cloud.

Motivating Scenario: We consider a Sensor-CloudInfrastructure (SCI) which is capable of providingenvironment monitoring services over an extensivegeographic region. End-users, such as weather fore-casting agencies, agricultural land owners, and in-dustrial plants, do not need to purchase and deployindividual WSNs. Instead, they obtain the informationof the environment condition in the form of Se-aaS.On the other hand, in order to provision Se-aaS, theSCSP relies on sensor nodes deployed by multiplesensor-owners in the concerned geographic region.Additionally, each sensor-owner depends on the othersensor-owners for transmitting sensed data from thesensor nodes to the BS via multihop communication,as mentioned earlier.

1.2 Contribution

In this paper, we introduce a Single Leader MultipleFollower Stackelberg game theory-based dynamic trustenforcing pricing (DETER) scheme for sensor-cloud,while considering that the sensor-owners form anoligopolistic market. In DETER, each sensor node inthe path, from the source node to the BS, selects theset of most suitable next-hop nodes. Thus, DETERensures an optimum trust value for the path andmaintains the QoS of Se-aaS provided by the SCSP.Additionally, the proposed scheme, DETER, ensuresthe profit of both the SCSP and the sensor-owners.We refer to the optimal strategies chosen by eachsensor node in the path obtained by using the DETERscheme as the Stackelberg equilibrium. In summary,the specific contributions of this work are given asfollows:

a) We propose a dynamic trust enforcing pricingscheme for service-oriented sensor-cloud, while con-sidering the possibility of selfish behavior by thesensor-owners.

b) In order to identify the misbehaving sensor-owners, we calculate distributed and centralized trustopinion sets for each owner using the beta reputationmodel proposed by Jøsang [7].

c) We use Single Leader Multiple Followers Stackelberggame theoretic approach for designing the proposedpricing scheme. In DETER, the SCSP acts as the leader,and the sensor-owners act as the followers.

d) We present four different algorithms to ensureQoS of Se-aaS, while choosing trust-enforced path fordata transmission in service-oriented sensor-cloud.

2 RELATED WORKS

In the existing literature, few works focus on theintegration of WSNs with the cloud, viz. [8], [9].However, the concept of sensor-cloud was conceivedby Yuriama et al. [9]. The authors [9] presented anoverview of the entire architecture of sensor-cloud.Further, the theoretical modeling of sensor-cloud,along with characterization of the concept of virtu-alization of sensor nodes used in sensor-cloud, waspresented by Misra et al. [1]. In another work, Boseet al. [10] studied the implementation of an infras-tructure, which is suitable for environmental moni-toring. Additionally, the practical implementation ofsensor-cloud was demonstrated by Madria et al. [11].Chatterjee et al. [12] studied an optimal compositionof virtual sensors, while considering the resource-constrained behavior of the nodes. The problem ofredundant data transmissions in sensor-cloud wasexplored by Chatterjee et al. [13], where a data cachingmechanism between the sensor nodes and the cloudwas suggested as a potential solution. In anotherwork, Chatterjee et al. [14] addressed the problem ofselecting the optimal data center using the optimaldecision rule. Misra et al. [15] presented an optimalduty scheduling algorithm suitable for sensor-cloudenvironment.

On the other hand, some of the existing worksconsider the movement of data from the source nodeto the BS in the sensor-cloud environment. Misraet al. [16] studied the selection of the optimal gate-way node for transmission of the sensed data to thecloud. In another work, Chatterjee et al. [17] pro-posed a scheme for selecting the optimal interme-diate node for data transmission, while consideringthe unintentional node failures. Therefore, we arguethat the problem of intentional misbehavior of sensornodes or sensor-owners in sensor-cloud has not beenexplored yet in the existing literature. There existseveral works in traditional WSNs, which considerthe presence of misbehavior of sensor nodes. Illianoet al. [18] surveyed different schemes proposed foranomaly detection and trust management. Li et al. [19]proposed a trust-based system for clustered WSNs.Rezvani et al. [20] studied a trust-based scheme forsecure aggregation of data using a modified iterative

3

filtering algorithm. In another work, Ahmed et al. [21]presented a-trust and energy-aware routing protocolfor WSNs, while considering hop counts and loadbalancing of routes. However, it is also not suitablefor a group of misbehaving nodes, which is a commonscenario in sensor-cloud. Rathore et al. [22] proposeda consensus aware socio-psychological trust modelto evaluate the trustworthiness of a node. However,these schemes are not suitable for using in the case ofsensor-cloud, because of the following reasons:

a) Most of these works consider cluster-based andlayered WSNs. The trust calculation process usedvaries based on cluster-heads and aggregators, whichare assumed to be trustworthy. However, these arenot suitable for sensor-cloud due to the presence ofmultiple sensor networks deployed by the differentsensor-owners.

b) Most of the proposed schemes deal with homoge-neous sensor nodes. However, sensor-cloud typicallydeals with heterogeneous sensor nodes, which areused for multitude of purposes.

c) Finally, in the existing literature, the proposedschemes aim to reduce the effects of the misbehaviorof nodes on the sensed data. However, there is a needto prevent these situations from occurring, which isone of the primary objectives of this work.

Therefore, in contrast to the previous works, agame theory based dynamic trust enforcement pricingscheme, DETER, for SOA of sensor-cloud is proposedin this paper in order to prevent competitive sensor-owners from misbehaving and ensure QoS. The pro-posed scheme, DETER, exploits the economic aspectsof the service-oriented sensor-cloud to encourage thesensor-owners to behave honestly.

3 SYSTEM MODELWe consider that the sensor-cloud has a few sensor-owners, as shown in Figure 1. Each sensor-ownersl ∈ S, where S is the set of the sensor-owners, isregistered with the SCSP. The set of heterogeneoussensor nodes owned by each sensor-owner sl is de-noted by Ol. In Figure 1, we denote heterogeneoussensor nodes using different geometrical shapes. Weconsider that the total set of sensor nodes registeredwith the SCSP is denoted by N . Hence, we have, N =O1∪O2∪· · ·∪O|S|. On the other hand, each end-user urregisters his/her service requirements with the SCSP,based on which a Service Level Agreement (SLA)[23], [24] is generated. The SLA states the servicedemanded by the user, and the agreed standards ofvarious parameters such as — the maximum tolerabledelay δ, and the maximum price Pmax that the end-user is willing to pay for the service. We consider thatthe SCSP follows the dynamic ’pay-per-use’ model,where the end-users pay based on their usage of theservice provided by the SCSP.

Based on the requirement of end-user ur, virtualsensors, denoted by vsj , are formed by the SCSP for

Fig. 1: Schematic Diagram of Sensor-Cloud

each application j using a subset of the physicalsensor nodes Nj ⊆ N , and allocated to ur. The setof virtual sensors assigned to each end-user ur isdenoted as vgr = {vs1, vs2, · · · , vsp}, where p is thetotal number of applications requested by the end-user ur. On the other hand, the selected source nodessense and transmit their sensed data to the SCSP usinga multi-hop path, as shown in Figure 2. The sensornodes use nodes belonging to other sensor-owners asintermediate node. Thereby, the sensor-owners of theintermediate selected nodes earn revenue for provid-ing services. Finally, the data is relayed to the cloud,where subsequent processing is performed to meet theend-user requirements.

TABLE 1: List of Symbols

Symbol DescriptionS Set of registered sensor-ownersOl Set of sensor nodes of sensor-owner slN Set of sensor nodes registered with SCSPvsj Set of virtual sensors for application jNj Set of physical sensors mapped to vsjvgr Set of virtual sensors assigned to user ur

Pmax Maximum price willing to pay by userϕ Profit earned by SCSPδ Maximum tolerable delay for a servicePt

pathg Utility function of SCSP for pathg

vtn,i Trust rating for node i by node nDOt

n,i Distributed trust opinion set for node iMt

n Candidate set decided by nrtn,i, h

tn,i Instantaneous feedback pair for node i

β Forgetting factorRt

n,i, Htn,i Cumulative feedback pair for node i

Ctn Centralized trust opinion set for node n

Dtscsp,i,n Discounted trust opinion setCOt

l Centralized trust opinion set for owner slUt

n,i(·) Utility function for node iEF t

n,i Expectation Factor for node ipti Pseudo-price for selecting next-hopTEt

l Trust expectation of sensor-owner sl

Assumptions: We list the assumptions consideredin the design of the proposed scheme, DETER.

• The SCSP, which is a centralized entity, is respon-sible for maintaining the dynamic pricing policy.

• The set of source nodes, which are selected bythe SCSP, transmit their sensed data to the cloudusing different paths. No aggregation of data isperformed by the intermediate nodes.

4

Fig. 2: Schematic Diagram of Sensor-Cloud with Het-erogeneous Sensor Nodes

• Each node has a list of sensor nodes belonging tothe same sensor-owner.

• Each node is capable of overhearing, while beingin the active or idle states [25], [26].

4 PROPOSED DYNAMIC TRUST ENFORCINGPRICING (DETER) SCHEME

The proposed scheme, DETER, comprises of two maincomponents — bi-level trust calculation, which is per-formed periodically by the sensor nodes and the SCSP,and dynamic pricing game, which is conducted at thebeginning of each service. The two components arediscussed in details in the subsequent sections.

4.1 Trust Function Calculation

In this work, we propose a bi-level trust factor calcu-lation method, which is motivated by the beta rep-utation model [7], and discounting and consensusoperators [27], [28]. We consider dual trust values,defined as the distributed trust opinion (DO) and thecentralized trust opinion (CO) sets. In the proposedscheme, DETER, at time instant t ∈ T , each sensornode n ∈ N calculates the DO set, where DOtn,i =(btn,i, d

tn,i, u

tn,i, a

tn,i). b

tn,i, d

tn,i, u

tn,i and atn,i are defined

as belief, disbelief, uncertainty, and atomicity, respec-tively, for the neighbor nodes, i.e., ∀i ∈ Nn ⊆ N ,where Nn denotes the set of neighbor nodes of sen-sor node n. It needs to satisfy the constraint —btn,i + dtn,i + utn,i = 1.

Eventually, these values become a key factor, fordeciding the candidate set (Mt

n) of potential data for-warding nodes, where Mt

n ⊆ Nn. On the other hand,the SCSP calculates the centralized trust opinion set,i.e., Ctn = (btscsp,n, d

tscsp,n, u

tscsp,n, a

tscsp,n), ∀n ∈ N ,

using the discounting and consensus operators, whileoperating on the previous trust opinion set Ct−1n =(bt−1scsp,n, d

t−1scsp,n, u

t−1scsp,n, a

tscsp,n), and the set of opinions

collected from the deployed sensor nodes. Here, wehave btscsp,n + dtscsp,n + utscsp,n = 1. In the followingsubsections, we discuss the method of calculation ofthe DO and CO sets.

4.1.1 Distributed Trust Opinion SetEach sensor node generates a DO set for each of itsneighbor nodes, periodically. In order to calculate theDO set, the sensor nodes notice the packet forwardingbehavior of their next-hop nodes by utilizing the prop-erty of overhearing. The steps of DO set calculationare listed as follows.

1) First, each node n calculates the trust rating vtn,ifor each node i ∈ Nn, as defined in Definition 1.

Definition 1. At time instant t, the trust rating vtn,icalculated by node n for node i is defined as the ratio of thenumber of correctly forwarded data packets F tn,i by node iamong those received from node n to the total number ofpackets T tn,i sent to node i by node n.

We assume that each node n has the informationwhether neighbor node i ∈ Nn belongs to the samesensor-owner as itself. If neighbor node i belongs tothe same sensor-owner, sk, as node n, trust rating isconsidered to be 1. Mathematically,

vtn,i =

{1, if {n, i} ∈ Ok and i ∈ NnF t

n,i

T tn,i, otherwise

(1)

2) Next, using the trust ratings, each node n de-termines the instantaneous positive and negative feed-back parameters [7] for each neighbor node i at timeinstant t, i.e., rtn,i and htn,i, respectively, using thefollowing equations:

(rtn,i, htn,i) =

(w(1 + vtn,i)

2,w(1− vtn,i)

2

)(2)

where w is defined by SCSP, and is a constant for noden, where w > 0.

3) Thereafter, the cumulative positive and negativefeedback pair (Rtn,i, H

tn,i) is calculated by node n for

each neighbor node i at time instant t, using thefollowing equations:

(Rtn,i, Htn,i) = (Rt−1n,i β + rtn,i, H

t−1n,i β + htn,i) (3)

where β is the forgetting factor, and is defined by theSCSP.

4) Finally, based on beta reputation model [7], eachnode n calculates DOtn,i for neighbor node i as fol-lows:

(btn,i, dtn,i, u

tn,i) =

(Rtn,i/K,H

tn,i/K, 2/K

)(4)

where K = (Rtn,i+Htn,i+2). Additionally, we consider

atn,i = 12 , so that, uncertainty is equally distributed

among belief and disbelief.

4.1.2 Centralized Trust Opinion SetThe SCSP calculates the CO set, Ctn, for each sensornode n ∈ N , i.e., Ctn = (btscsp,n, d

tscsp,n, u

tscsp,n, a

tscsp,n),

using the consensus and discounting operators [27]–[29] on the DO sets obtained from each node. There-after, the SCSP calculates the CO set for each sensor-

5

owner sl ∈ S, COtl , based on the calculated CO setof each of its sensor node n ∈ Ol. The steps forcalculation of CO set for each sensor-owner by theSCSP are listed as follows:

1) Initially, the SCSP defines the CO set for each sen-sor node n ∈ N at time instant t0 as Ct0n = (1, 0, 0, 0.5).Thereafter, it collects the DO sets from the individualsensor node i for each of its neighbor nodes, n ∈ Ni,periodically.

2) Next, the SCSP calculates the discounted opinionof node n, i.e., Dtscsp,i,n, using its own opinion of nodei in the previous time instant, Ct−1i and the DO sets.Hence, Dtscsp,i,n is determined as follows1:

Dtscsp,i,n = Ct−1i ⊗DOti,n (5)

It should be noted that the discounted opinions calcu-lated by the SCSP are considered to be independent.

3) The discounted opinion values are then, com-bined by the SCSP using the consensus operator toobtain the CO set for each node n, which is calculatedas follows1:

Ctn = Dtscsp,i,n ⊕ · · · ⊕ Dtscsp,j,n (6)

where Ctn = (btscsp,n, dtscsp,n, u

tscsp,n, a

tscsp,n) and nodes

{i, · · · , j} refer to those nodes in the system which donot belong to sensor-owner as node n.

4) Finally, the CO set1, COtl = (Btl , Dtl , U

tl , A

tl), for

each sensor-owner sl is calculated by SCSP as theaverage of Ctn values of all nodes belonging to thesensor-owner sl.

With the help of these trust values, DETER enforcestrust among oligopolistic sensor-owners in sensorcloud using a game-theoretic dynamic pricing scheme,which is discussed in the following subsection.

4.2 Game-Theoretic Dynamic Pricing SchemeIn this work, we model the interaction between theSCSP and a few registered sensor-owners in service-oriented sensor-cloud using a Single Leader MultipleFollower Stackelberg game. Here, we consider that theSCSP acts as the leader, and decides the price to bepaid to each sensor-owner, while ensuring QoS of thenetwork and maximizing its own profit. On the otherhand, the sensor-owners act as the followers, andprovide the service as requested by the SCSP, whileensuring high revenue. The justification for usingStackelberg game theory in this work is discussed asfollows.

4.2.1 Stackelberg Game: The Justification:Game theory is a mathematical tool used to studythe complexities of decision making for individu-als in competitive scenarios, where each individualinfluences the other individuals. As mentioned inSection 1, in the SOA of sensor-cloud, there exists

1. For detailed calculation, refer to supplementary file.

an oligopolistic market scenario among a few non-cooperative sensor-owners, where each sensor-ownercompetes with the other sensor-owners to earn higherprofits. Thus, some of the sensor-owners may adoptunfair means to degrade the market reputation ofothers. These interactions among the sensor-owners,and the sensor-owners and the SCSP are modeledusing game theoretic approach. Furthermore, sincesensor-cloud has a SOA, it is highly essential that thequality of Se-aaS provided by the SCSP is compliantwith the QoS specified in the SLA. Hence, in orderto ensure the specified QoS, the SCSP needs to adoptmeasures for preventing the competitive oligopolisticmarket situation among the sensor-owners. Therefore,we use a Single-Leader-Multiple-Follower StackelbergGame in the proposed scheme, DETER. In DETER, theSCSP acts as the leader and takes the pricing decision,while the sensor-owners act as followers and takethe decision to behave honestly or dishonestly in theservice-oriented sensor-cloud.

4.2.2 Game FormulationIn DETER, initially, the end-users request services tothe SCSP, as per their requirements. We consider thatthe end-user provides information about the type ofservice required, and the tolerable delay, δ. Based onthese information, the SCSP decides the price to becharged, which is denoted by Pmax. The price Pmax isfixed for specific service as per the agreement betweenthe SCSP and the end-user. In the following sections,we discuss the utility functions of the proposed Stack-elberg game along with the equilibrium conditionsand the solution of the game.

Utility Function for Next-Hop SelectionOn receiving the service specifications from the end-users, the SCSP selects the source sensor nodes fordelivering the requested services. Each source nodeneeds to forward the generated information to theSCSP within the maximum tolerable delay specifiedby the end-user. Since, in sensor-cloud, multi-hopcommunication is used for forwarding data to the BS,the source sensor nodes take help of the intermediatesensor nodes, which may belong to any sensor-owner.In order to select the optimum trustworthy hop nodesfor data transmission, each node n calculates a utilityfunction, U tn,i(·), for each neighbor node i at timeinstant t and tries to maximize its payoff.

Motivated by the work of Chatterjee et al. [2], weconsider that each sensor node n calculates payoffof U tn,i(·) for the neighbor nodes at each step, anddecides the candidate set of potential data forwardingnodes, Mt

n. The different parameters of U tn,i(·) arediscussed as follows:

Residual Energy Factor (ρtn,i)1: We define the resid-

ual energy factor of node i at time instant t, ρtn,i, as theratio of the residual energy of node i at time instant t,Etres,i, and its initial energy, Einit,i. Hence, we have,

6

U tn,i(·) > U tn,j(·) and i � j, where (ρtn,i > ρtn,i),{i, j} ∈ Nn, and other parameters are constant.

Received Signal Strength (RSStn,i)2 : We define

the received signal strength, RSStn,i, using the Friis’transmission formula [30], assuming that there is nopower loss due to hardware [31]. With the increasein RSStn,i, the payoff of U tn,i(·) increases.

Signal-to-Noise Ratio (SNRtn,i)2: The signal to noise

ratio, SNRtn,i, is formulated based on the work ofEtezadi et al. [32]. With the increase in RSStn,i, thepayoff of U tn,i(·) increases.

Distributed Trust Opinion Set (DOtn,i): We get thedistributed trust opinion set, DOtn,i, using Equation(4). We consider that U tn,i varies proportionally withthe expectation factor, which is defined in Definition 2.

Definition 2. Expectation factor, EFtn,i, is the ratio of thetrust expectation value, BEtn,i, and the distrust expectationvalue, DEtn,i. Here, BEtn,i = btn,i + atn,iu

tn,i, and DEtn,i =

dtn,i + (1 − atn,i)utn,i, and atn,i determines the fraction ofuncertainty that can contribute as belief.

By choosing a next-hop node i with high valuesof the above mentioned factors, the node n ensuresthat high quality data is provided to the SCSP, i.e., itgains in terms of QoS. We quantify the gain of noden, for selecting node i, in terms of the virtual revenuefunction, RF tn,i(·), as follows,

RF tn,i(·) = ω1ρtn,i+ω2

RSStn,iRSSmax

+ω3

SNRtn,iSNRmax

+ω4EFtn,i

(7)where ω1, ω2, ω3 and ω4 are constants representingthe effect of each parameter on the revenue func-tion. RSSmax and SNRmax define the received signalstrength and the signal-to-noise ratio at distance d0,where 0 < d0 << rn, and rn is the communicationrange of node n.

Pseudo-Price for Selecting Next-Hop (pti): The pricepti to be paid to the selected next-hop node i ∈ Ol hasa negative impact over U tn,i(·), where pti is defined asfollows:

pti = ΦTEtl2 (8)

where Φ is a constant, and TEtl is the trust expectationvalue for the sensor-owner sl, where i ∈ Ol, which iscalculated by SCSP and defined as — TEtl = Btl+A

tlU

tl .

Distance from the Base Station (dBS,i): The utilityfunction, U tn,i(·), varies inversely with the distancebetween the node i and the BS, dBS,i, as selecting ahop node closer to the BS reduces the total numberof hops in the path.

Here, both the pseudo-price function as well as thedistance from BS of the node i have a negative impacton the payoff from the utility function. Hence, weconsider these two factors as the cost node n bearsfor choosing node i as the next-hop and define the

2. For detailed calculation, refer to supplementary file.

virtual cost function, CF tn,i(·), as follows:

CF tn,i(·) = ω5pti + ω6

dBS,idBS,n

(9)

where ω5 and ω6 are constants.Therefore, we define the utility function U tn,i(·), of

node i evaluated by node n, as the net virtual profitthat node n incurs for choosing node i as the next-hop. We have,

U tn,i(·) = RFtn,i(·)− CF

tn,i(·) (10)

Each node n tries to maximizes the payoff of U tn,i(·)for choosing optimal candidate set, Mt

n, while satis-fying the following constraints along with RSStn,i ≥RSSth,SNRtn,i ≥ SNRth:

Etres,i ≥ Eres,th, dBS,i ≤ dBS,n (11)

where Eres,th and bth indicate the threshold valuesfor residual energy and belief of a node, respectively.RSSth and SNRth define the received signal strengthand the SNR at distance rn, where rn is the commu-nication range of node n. Hence, after calculating thepayoff values for each node i ∈ Nn using Equation(10), the sensor node n decides the candidate set ofpotential data forwarding nodes, Mt

n, where |Mtn| is

decided by the SCSP.

Utility Function for Path SelectionThe utility function of the SCSP, Ptpathg

, is definedas the revenue of the SCSP for selecting path pathg ,while taking into account the trust value of the chosenpath pathg . We consider that Npathg

defines the set ofsensor nodes belonging to the path pathg . Given themaximum tolerable delay, δ, the following constraintneeds to be satisfied for each service:

|Npathg | ≤ (δ + 1) (12)

We consider that unit delay is incurred at each hop,and there are |Npathg

− 2| number of intermediatesensor nodes. When the control packets forwardedby the source sensor nodes reach the BS throughmultiple paths, the SCSP calculates the payoff ofthe utility function for each path pathg , and selectsthe path having the maximum payoff. We considerthat the price Pmax, to be paid by the end-users,includes the profit of the SCSP with two types ofcost — hardware cost, Phw, and infrastructure cost,Pinfra [2], i.e., Pmax = Phw +Pinfra+ϕ. We considerthat Pinfra is directly proportional to the number ofvirtual sensors used for a service, and the duration ofservice. For a particular virtual sensor, the price to bepaid is constant for unit time. On the other hand, therevenue of the SCSP, ϕ, varies based on the trust factorof the selected sensor-owners. In order to ensure non-negative profit of the SCSP, we need to ensure ϕ > 0.Phw depends on the cost incurred due to the use ofsensor nodes for providing a particular service, and

7

Phw = Pshw+Prhw. Pshw and Prhw are the costs incurredfor sensing and relaying, respectively. Pshw is only paidto the source nodes, which generate the sensor data.Prhw =

∑∀n Pn, and n ∈ pathg . Pn defines the price

to be paid to each sensor node n, which are in pathpathg . We assume that Pn for sensor node n variesproportionally with the normalized pseudo price, TEtk,of the sensor-owner k, where n ∈ Ok. Hence, we get:

Pn = χ[TEtk2/∑

n∈pathg

{TEtk2|n ∈ Ok}] (13)

where χ is a constant and defines the price for fullytrusted sensor-owner sk, i.e., COtk = (1, 0, 0, Atk).Hence, for provisioning service within the specifieddelay and ensuring high profit, the SCSP tries tomaximize the payoff of the utility function, Ptpathg

,defined as below:

Ptpathg= 1− Phw

Pmax+

n∈pathg∏k,n∈Ok

Btk +AtkUtk

Dtk + (1−Atk)U tk

(14)

To account for the cumulative effect of the trustopinion set of each node in path pathg , we introducethe product function in Equation (14).

4.2.3 Existence of Stackelberg EquilibriumAs discussed in Section 4.2.2, in DETER, the selectionof the next-hop node i by node n not only depends onthe physical parameters of node i, but also dependson the distributed trust opinion of node n for nodei and the centralized trust opinion of SCSP for theowner of node i. Additionally, the price to be paid bythe SCSP to node n depends on the action taken bynode n, i.e., the next hop selected by node n, andthe centralized trust opinion for owner of node n.Thereby, there exists a non-cooperative game amongthe competitive sensor-owners, in which each sensor-owner tries to maximize his/her revenue by increas-ing the chances of selection of his/her sensor nodesas well as by selecting the optimum next-hop nodefor data forwarding.

Given the values of the centralized trust opinionsets calculated by the SCSP, the optimal response,known as the reaction set of the sensor nodes, isdetermined as the set of Nash Equilibrium strategiesfor the sensor nodes. Here, Nash Equilibrium refersto the equilibrium state of the followers at whichnone of the players, i.e., sensor owner, gain benefit byunilaterally deviating from the chosen strategy [33]On obtaining the reaction set of the sensor nodes,the SCSP decides the optimum strategy, i.e., the pathfor data-transmission, having maximum payoff value,which is known as Stackelberg Equilibrium [34]. Inother words, Stackelberg Equilibrium refers to theequilibrium state of the system having leader-followerhierarchy at which the leader chooses the strategywhich yields the maximum pay-off given the set of

Nash Equilibrium strategy of the followers. Hence,we define the Stackelberg equilibrium for the non-cooperative oligopolistic market of sensor-cloud asin Definition 3. Additionally, the existence of Nashequilibrium is shown in Theorem 1. For the dis-tributed equilibrium solution of DETER, the readeris requested to refer to the supplementary file.

Definition 3. The Stackelberg equilibrium of DETER isdefined as (bt∗n,i, d

t∗n,i, B

t∗l , D

t∗l ), where the proposed scheme

ensures the following conditions are satisfied:

U tn,i(bt∗n,i, dt∗n,i, Btl , Dtl ) ≥ U tn,i(btn,i, dtn,i, Btl , Dt

l ) andPtpathg

(bt∗n,i, dt∗n,i, B

t∗l , D

t∗l ) ≥ Ptpathg

(bt∗n,i, dt∗n,i, B

tl , , D

tl )

(15)where bt∗n,i, d

t∗n,i are the optimum belief and disbelief values

of the chosen node i ∈ Nn by node n, and Bt∗l , Dt∗l are

the optimum belief and disbelief values of sensor-owner l,where i ∈ Ol.

Theorem 1. Given the set of neighbor nodes Nn of noden, and the price function, which varies polynomially withthe trust expectation value TEtl , where ∀i ∈ (Nn ∩ Ol),there exists Nash equilibrium, if we have negative valuesfor variational inequality of U tn,i(·) with respect to btn,i,and positive values for variational inequality of U tn,i(·) withrespect to dtn,i.

Proof: For the proof of Theorem 1, refer to sup-plementary file.

Lemma 1. From Theorem 1, we conclude that DETERensures the existence of Stackelberg Equilibrium, as thereexists Nash Equilibrium among the followers. The Stack-elberg Equilibrium can be obtained by selecting the pathwhich maximizes the payoff of the utility function Ptpathg

of the SCSP.

5 PROPOSED ALGORITHMSThe detailed working of the proposed scheme DETERis presented using an interaction diagram in Figure 3.In DETER, each sensor node n calculates DO set forits neighbor nodes, Nn, periodically, using Algorithm1. Based on DO calculated by the sensor nodes, N ,the SCSP calculates the CO for the registered sensor-owners, S, using Algorithm 2. In sensor-cloud, thesource nodes needs to establish a path to BS, in orderto transmit sensed data to the SCSP. Hence, the sourcenode and the selected intermediate nodes executeAlgorithm 3 to determine the candidate set for next-hop node selection. Thereafter, the SCSP performs anoptimization algorithm, i.e., Algorithm 4, for selectingthe path having optimum cumulative trust valuesto provide high quality service. Using Algorithm 4,the SCSP also ensures its own profit and the profitof the sensor-owners based on the their centralizedtrust opinion sets. The detailed pseudo-codes of theAlgorithms 1-4 are given in the supplementary file.

8

(a) Sensor-Cloud (b) Trust Calculation

Fig. 3: Interaction Diagram of DETER

1

BeliefValue

of a Node

0.80.6

0.40.2

00

Belief Value of a Owner

0.2

0.4

0.6

0.8

15

50

45

40

35

30

25

20

1

Utility

Value

20

25

30

35

40

45

Fig. 4: Theoretical Analysis of Utility value

1

Disbelief

Value of a

Owner

0.80.6

0.40.2

00

Belief Value of a Owner

0.2

0.4

0.6

0.8

0.6

0.4

1

0.2

0

0.8

1

Norm

alizedPrice

Earned

bytheOwner

Fig. 5: Price with Varying Trust of Sensor-Owners

6 PERFORMANCE EVALUATION6.1 Simulation ParametersTo evaluate the performance of the proposed scheme,DETER, we consider that the heterogeneous sensornodes, i.e., having different types of sensors, are de-ployed over the terrain in a random pattern, andthe BS is at the center of the terrain, in a MATLABsimulation platform. Additionally, we consider thatthe heterogeneous sensor nodes use the IEEE 802.15.4protocol for communication, as shown in Table 2.Hence, the size of Hello packet is 6 bytes, andeach Data packet has a payload of 128 bytes. Forsimulation purpose, we have selected the source noderandomly based on the available sensors and theapplication requirements of the end-users. Initially, weconsider that each node sends 100 packets to its neigh-bor nodes for calculating the distributed trust opinionsets. Thereafter, based on the request of the SCSP, thesensor nodes send packets to the BS. Additionally, weassume that for each service of one hour, 180 packetsare sent to the SCSP, as shown in Table 2.

6.2 BenchmarksWe compared the proposed scheme, DETER, with twodifferent existing schemes — dynamic optimal pric-

ing for heterogeneous services-oriented architecture(referred in this paper as DOP) [2] and trust andenergy aware routing protocol (TERP) [21]. In DOP[2], Chatterjee et al. proposed a pricing scheme forhardware resource usage in sensor-cloud. The authorsproposed that path selection is done at the beginningof each service based on physical parameters of nodes.The price charged by the owners of the selected nodesis determined from the price charged by the previoushop and the estimated price to be paid by the end-user. However, the selfish behavior of the sensor-owners is not considered in DOP. Hence, trust of theselected path is not ensured. On the other hand, inTERP, Ahmed et al. [21] studied a routing schemeto eliminate misbehaving or faulty nodes from path.Trust rating of each node is determined based onits opinion about itself as well as the opinions ofits neighbor nodes. However, this approach cannotbe used in sensor-cloud, as the presence of multi-ple sensor-owners and the economic aspect of thesensor-cloud, i.e., the profits of the SCSP and thesensor-owners, are not considered. Using the pro-posed scheme, DETER, we can ensure QoS of Se-aaSwith high trust values in presence of multiple sensor-owners and high profits, than using DOP and TERP.

9

0.90

0.94

0.981

0 10 20 30 40 50

Bel

ief

Val

ue

Time (Hour)

0.90

0.94

0.981

0 10 20 30 40 50

Time (Hour)

0.90

0.94

0.981

0 10 20 30 40 50

Time (Hour)

0.22

0.24

0.26

0.28

0 10 20 30 40 50

Pric

e E

arne

d (i

n un

its)

Time (Hour)

DETER

Owner-1

0.22

0.24

0.26

0.28

0 10 20 30 40 50

Time (Hour)

DOP

Owner-2 Owner-3

0.22

0.24

0.26

0.28

0 10 20 30 40 50

Time (Hour)

TERP

Owner-4

Fig. 6: Comparitive Analysis of Belief Value and Price Earned by Sensor-Owners

TABLE 2: Simulation Parameters

Parameter ValueSimulation area 1000 m×1000 mNumber of sensor nodes 400− 800Number of sensor owners 4Type of sensor nodes 3

|Mtn| 3− 5

Number of BS 1Communication protocol IEEE 802.15.4Initial energy of each node 20 J [31]Communication range 100 mPacket Header size 6 bytesPacket Payload size 128 bytesPacket transmission rate 180 packets/hr/serviceTx energy consumption 50 nJ/bit [35]Rx energy consumption 50 nJ/bit [35]Energy consumption at amplifier 100 pJ/bit-m2 [35]

6.3 Performance MetricsWe evaluated the performance of the proposedscheme, DETER, using the following metrics.

Explored Nodes: To provide a particular service foran hour, the value of explored nodes is calculated asthe average of the total number of nodes exploredin the network, before finalizing the path, from thechosen source node to the BS. With the increase in thenumber of explored nodes in the network, the systemperformance deteriorates.

Total and Correctly Forwarded Packets: We calculatethe network overhead by analyzing the total numberof data packets sent by the nodes. For evaluation, weconsidered the average number of packets forwardedby the nodes in an hour as the total forwarded packets,totPackets. Additionally, the percentage of correctlyforwarded packets, percentCorPackets, varies lin-early with the ratio of the number of packets for-warded correctly, corPackets, to the total numberof packets forwarded.

Average Path Length: The average number of nodesin the path selected for each hour of service is termedas the average path length, Npathg

. Hence, the averagenumber of hops in each path is given by, (Npathg

−1).Average Number of Selfish Nodes in Path: We consider

a node as selfish, if it belongs to a selfish sensor-owner. With the increase in the number of selfishnodes in a path, the QoS provided by the SCSP tothe end-user decreases.

Network Lifetime: We define network lifetime as thetime elapsed after node deployment till when the lastnode dies in the network. For better performance ofthe network, we need to ensure high network lifetime.

Price paid by SCSP to sensor-owners: The profits ofthe sensor-owner increases linearly with the increasein the price paid to them by the SCSP. Each sensor-owner sl ∈ S gets paid, if and only if, any node ibelonging to the sensor-owner, i.e., i ∈ Ol, is presentin the selected path, pathg .

Profit of SCSP: The profit of the SCSP is defined asthe difference between the amount received by theSCSP from the end-users in exchange of the providedservice and the amount spent by the SCSP for provi-sioning the service, which includes the price paid tothe sensor-owners and the cost incurred for providinginfrastructural resources.

6.4 Results and Discussions

We observe that, in DETER, with the increase in beliefvalue of each sensor-owner sl, the utility value of eachnode n ∈ Ol varies almost exponentially, consideringthat the belief of node n is fixed, as shown in Figure4. On the other hand, with the increase in the beliefvalue of each sensor node n ∈ Ol, the utility value ofthe node varies insignificantly with the belief valueof the sensor-owner remaining fixed. This is becausein DETER, the next hop node selection is highlydependent on the belief value of the sensor-owners,in order to resist monopoly situation. In Figure 5,we observe that for fixed belief value, the price paidby the SCSP to a sensor-owner decreases with theincrease in his/her disbelief value. With the increasein belief value and decrease in disbelief value of thesensor-owner, s/he earns higher price from SCSP.

10

0

200

400

600

800

1000

400 600 800

Ave

rage

Num

ber

of N

odes

Exp

lore

d f

or E

ach

Serv

ice

(per

hou

r)

Number of Nodes

DETER DOP TERP

0

20

40

60

80

400 600 800

Fig. 7: Number of Nodes Explored

0

40000

80000

120000

160000

200000

400 600 800

Tot

al P

acke

ts F

orw

arde

d (

Per

Hou

r)

Number of Nodes

DETER DOP TERP

Fig. 8: Network Overhead

84

88

92

96

400 600 800

Perc

enta

ge o

f co

rrec

tly F

orw

arde

d P

acke

ts (

Per

Hou

r)

Number of Nodes

DETER DOP TERP

Fig. 9: Correctly Forwarded Packet

0.88

0.9

0.92

0.94

0.96

0.98

1

400 600 800

Nor

mal

ized

Res

idua

l Ene

rgy

(per

nod

e)

Number of Nodes

DETER DOP TERP

Fig. 10: Residual Energy of Nodes

0

20

40

60

80

400 600 800

Net

wor

k L

ifet

ime

(in

sim

ulat

ion

days

)

Number of Nodes

DETER DOP TERP

Fig. 11: Network lifetime

0

10

20

30

400 600 800

Ave

rage

Pat

h L

engt

h

Number of Nodes

DETER DOP TERP

Fig. 12: Average Path Length

Moreover, a comparative analysis of DETER, DOPand TERP is shown in Figure 6. We observe that thebelief value of each sensor-owner increases with timeusing DETER, whereas for DOP and TERP, it remainsconstant with time. This is due to the fact that DETERencourages each sensor-owner to behave honestly bypaying less price as shown in Figure 6. Additionally,we observe that the price earned by each sensor-owner increases with the increase in individual beliefvalue. However, in DOP, the sensor owners earn thesame price irrespective of their belief values, as DOPdoes not consider trust value for price calculation. Incase of TERP, the price earned by each sensor-ownerdepends on his/her belief value. As the belief valuesremain fixed, the price earned by the sensor-ownersremain unchanged.

Figures 7, 8, and 9 depict that DETER outperformsthe existing schemes — DOP and TERP. Though theaverage number of nodes explored in DETER is higherthan that in DOP, DETER explores 76.57-86.47% lessnumber of nodes than TERP. In the path explorationphase, unlike DOP, DETER and TERP explore mul-tiple paths from each source node to the BS. TheSCSP finalizes the path to be used for data forwardingfrom the multiple choices of paths. However, DOPdepends on single path exploration for each source-BS pair. Moreover, DETER and TERP are capableof facilitating the change of path, if any faulty ormisbehaving node is detected in the path. We alsoget that the network overhead decreases by 52.6-56.53% using DETER, than using TERP. Additionally,using DETER, the percentage of correctly forwardedpackets increases by 0.23-0.62% and 0.27-0.56% thanusing DOP and TERP, respectively. Additionally, weobserved that in sensor-cloud, TERP does not alwaysguarantee a path between the source node and the

BS, as in TERP, each node only chooses the next-hopnodes which have trust value greater than 0.6 [21].

Using DETER, the residual energy of each node isalmost similar to that using DOP, and 4.69-11.56%higher than that of TERP, as depicted in Figure 10.This is due to the fact that, using DOP, the energyconsumed for node exploration is less than usingDETER and TERP. Therefore, we conclude that DE-TER ensures trust-based QoS in sensor-cloud withoutconsuming high amount of energy. The claim is alsosupported by the Figure 11. From Figure 11, weobserve that the network lifetime is higher using DOPthan using DETER, as using DOP, the number ofnodes explored increases linearly with the increasein hop count. Whereas, using DETER and TERP, thethe number of nodes explored increases exponentiallywith the increase in hop count. In DETER, owing tothe restriction in the cardinality of the candidate set,the number of nodes explored is bounded. However,using TERP, the available nodes having trust ratinggreater than or equal to 0.6 are explored. Therefore,using TERP, the energy consumption of the networkis higher than using DETER. Hence, we observe thatthere is a significant increase in the network lifetimeby using DETER as compared to TERP.

In Figure 12, we observe that for a service, the av-erage length of the final path obtained using DETERimproves by 10.87-23.21% than using TERP. UsingTERP, the average path length is higher than DETERdue to the fact that in TERP, the intermediate nodesin a path must have trust values greater than 0.6[21], as mentioned earlier. In DOP, neighbor nodeswhich are closer to the BS are given higher preference.Hence, with the increase in number of total deployednodes, the average path length shows a decreasingtrend. However, since trust value of the nodes is not

11

1

2

3

4

5

6

400 600 800Ave

rage

Num

ber

of S

elfi

sh N

odes

in P

ath

Number of Nodes

DETER DOP TERP

Fig. 13: Selfish Nodes in a Path

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

400 600 800

Ave

rage

Pri

ce P

aid

by S

CSP

to

Sen

sor

Ow

ners

(in

uni

ts)

Number of Nodes

DETER DOP TERP

Fig. 14: Price Paid by SCSP

700

800

900

1000

1100

1200

1300

1400

1500

1600

400 600 800

Ave

rage

Pro

fit o

f SC

SP (

in u

nits

)

Number of Nodes

DETER DOP TERP

Fig. 15: Profit of the SCSP

considered, DOP fails to ensure trust of the path. Onthe other hand, using DETER, the number of selfishnodes in the path is reduced by 35.61-52.72% and26.43-61.37% than using DOP and TERP, respectively,as shown in Figure 13.

Figures 14 and 15 imply that using DETER, theamount earned by the sensor-owners and the profitof the SCSP is almost similar with the increase inthe number of nodes. However, using the existingschemes — DOP and TERP, high profit of the sensor-owners and the SCSP cannot be ensured at the sametime, in sensor-cloud. From Figure 14, we note thatthe price to be paid by the SCSP to the sensor-ownersdecreases by at most 25.2% and 30% by using DETER,than using DOP and TERP, respectively. Correspond-ingly, we observe that the profit of the SCSP increaseby at most 50.15% and 73.9% using DETER than byusing DOP and TERP, respectively. We obtain theseresults due to the fact that unlike DETER, the SCSPdoes not penalize the selfish sensor owners in DOP.Thus, using DOP, the profit of the SCSP reducesthan using DETER. Therefore, we conclude that inDETER, we ensure trust in the presence of multiplesensor-owners in sensor-cloud by using the proposeddynamic pricing scheme. Moreover, DETER yieldshigh network lifetime, while ensuring high QoS insensor-cloud as compared to the existing schemes.

7 CONCLUSION

In this paper, we formulated a Single-Leader-Multiple-Follower Stackelberg game theory-based dy-namic pricing scheme for sensor-cloud in order toenforce trust among the oligopolistic sensor-owners.In DETER, each sensor node calculates distributedtrust opinion set for its neighbor nodes, locally. Basedon the distributed trust opinion sets, the SCSP de-termines the centralized trust opinion set for eachregistered sensor-owner. The sensor nodes use thesetrust opinion sets to select their next-hop nodes. Onthe other hand, the SCSP decides the price to bepaid to each sensor-owner based on the centralizedtrust opinion set. From simulation, we observe thatusing DETER, network lifetime and the percentage ofcorrectly forwarded packets increases. Moreover, thenumber of selfish nodes in the path decreases usingDETER which compels the sensor-owners to behave

honestly. Additionally, DETER ensures high profits ofboth the sensor-owners and the SCSP.

This work can be extended to ensure high QoSof Se-aaS provided by the SCSP in the presence ofunintentional failures of the sensor nodes in sensor-cloud. Additionally, the influence of external attackersin sensor-cloud can also be explored. It can also beextended to understand service provisioning mecha-nism for mobile sensor-cloud in presence of multiplesensor-owners.

REFERENCES

[1] S. Misra, S. Chatterjee, and M. S. Obaidat, “On TheoreticalModeling of Sensor Cloud: A Paradigm Shift From WirelessSensor Network,” IEEE Systems Journal, vol. 11, no. 2, pp. 1084–1093, Jun 2017.

[2] S. Chatterjee, R. Ladia, and S. Misra, “Dynamic OptimalPricing for Heterogeneous Service-Oriented Architecture ofSensor-cloud Infrastructure,” IEEE Trans. on Services Comput-ing, vol. 10, no. 2, pp. 203–216, Mar 2017.

[3] P. Chavali and A. Nehorai, “Managing Multi-Modal SensorNetworks Using Price Theory,” IEEE Trans. on Signal Process-ing, vol. 60, no. 9, pp. 4874–4887, Sept 2012.

[4] L. Guijarro, V. Pla, J. R. Vidal, and M. Naldi, “Maximum-ProfitTwo-Sided Pricing in Service Platforms Based on WirelessSensor Networks,” IEEE Wireless Communications Letters, vol. 5,no. 1, pp. 8–11, Feb 2016.

[5] K. W. Park, J. Han, J. Chung, and K. H. Park, “THEMIS: AMutually Verifiable Billing System for the Cloud ComputingEnvironment,” IEEE Trans. on Services Computing, vol. 6, no. 3,pp. 300–313, Jul 2013.

[6] S. Chaisiri, B. S. Lee, and D. Niyato, “Optimization of Re-source Provisioning Cost in Cloud Computing,” IEEE Trans.on Services Computing, vol. 5, no. 2, pp. 164–177, Apr 2012.

[7] A. Jøsang and R. Ismail, “The Beta Reputation System,” inProc. of the 15th Bled Electronic Commerce Conf., vol. 5, 2002,pp. 2502–2511.

[8] K. Ahmed and M. Gregory, “Integrating Wireless Sensor Net-works with Cloud Computing,” in Proc. of the 7th Int. Conf. onMobile Ad-hoc and Sensor Networks, Dec 2011, pp. 364–366.

[9] M. Yuriyama and T. Kushida, “Sensor-Cloud Infrastructure- Physical Sensor Management with Virtualized Sensors onCloud Computing,” in Proc. of the 13th Int. Conf. on Network-Based Information Systems, Sept 2010, pp. 1–8.

[10] S. Bose, N. Mukherjee, and S. Mistry, “Environment Monitor-ing in Smart Cities Using Virtual Sensors,” in Proc. of the 4th

IEEE Int. Conf. on Future Internet of Things and Cloud, Aug 2016,pp. 399–404.

[11] S. Madria, V. Kumar, and R. Dalvi, “Sensor Cloud: A Cloud ofVirtual Sensors,” IEEE Software, vol. 31, no. 2, pp. 70–77, Mar2014.

[12] S. Chatterjee and S. Misra, “Optimal Composition of a VirtualSensor for Efficient Virtualization within Sensor-Cloud,” inProc. of IEEE Int. Conf. on Communications, Jun 2015, pp. 448–453.

12

[13] S. Chatterjee and S. Misra, “Dynamic and Adaptive DataCaching Mechanism for Virtualization within Sensor-Cloud,”in Proc. of IEEE Int. Conf. on Advanced Networks and Telecom-muncations Systems, Dec 2014, pp. 1–6.

[14] S. Chatterjee, S. Misra, and S. Khan, “Optimal Data CenterScheduling for Quality of Service Management in Sensor-Cloud,” IEEE Transactions on Cloud Computing, 2015, DOI:10.1109/TCC.2015.2487973.

[15] T. Ojha, S. Bera, S. Misra, and N. S. Raghuwanshi, “DynamicDuty Scheduling for Green Sensor-Cloud Applications,” inProc. of the IEEE 6th Int. Conf. on Cloud Computing Technologyand Science, Dec 2014, pp. 841–846.

[16] S. Misra, S. Bera, A. Mondal, R. Tirkey, H. C. Chao, andS. Chattopadhyay, “Optimal Gateway Selection in Sensor-Cloud Framework for Health Monitoring,” IET Wireless SensorSystems, vol. 4, no. 2, pp. 61–68, Jun 2014.

[17] S. Chatterjee, S. Sarkar, and S. Misra, “Energy-Efficient DataTransmission in Sensor-Cloud,” in Proc. of Applications andInnovations in Mobile Computing, Feb 2015, pp. 68–73.

[18] V. P. Illiano and E. C. Lupu, “Detecting Malicious Data Injec-tions in Wireless Sensor Networks: A Survey,” ACM ComputingSurveys, vol. 48, no. 2, pp. 24:1–24:33, 2015.

[19] X. Li, F. Zhou, and J. Du, “LDTS: A Lightweight and Depend-able Trust System for Clustered Wireless Sensor Networks,”IEEE Trans. on Information Forensics and Security, vol. 8, no. 6,pp. 924–935, Jun 2013.

[20] M. Rezvani, A. Ignjatovic, E. Bertino, and S. Jha, “Secure DataAggregation Technique for Wireless Sensor Networks in thePresence of Collusion Attacks,” IEEE Trans. on Dependable andSecure Computing, vol. 12, no. 1, pp. 98–110, Jan 2015.

[21] A. Ahmed, K. A. Bakar, M. I. Channa, K. Haseeb, and A. W.Khan, “TERP: A Trust and Energy Aware Routing Protocol forWireless Sensor Network,” IEEE Sensors Journal, vol. 15, no. 12,pp. 6962–6972, Dec 2015.

[22] H. Rathore, V. Badarla, and S. Shit, “Consensus-Aware So-ciopsychological Trust Model for Wireless Sensor Networks,”ACM Trans. on Sensor Networks, vol. 12, no. 3, pp. 21:1–21:27,Aug 2016.

[23] D. Ardagna, B. Panicucci, and M. Passacantando, “GeneralizedNash Equilibria for the Service Provisioning Problem in CloudSystems,” IEEE Trans. on Services Computing, vol. 6, no. 4, pp.429–442, Oct 2013.

[24] D. Ardagna, M. Ciavotta, and M. Passacantando, “GeneralizedNash Equilibria for the Service Provisioning Problem in Multi-Cloud Systems,” IEEE Trans. on Services Computing, vol. 10,no. 3, pp. 381–395, May 2017.

[25] S. Buchegger and J.-Y. Boudec, “Performance Analysis of theCONFIDANT Protocol,” in Proc. of ACM International Sympo-sium on Mobile Ad-Hoc Networking & Computing, New York,USA, 2002, pp. 226–236.

[26] U. B. I. Maarouf and A. R. Naseer, “Efficient MonitoringApproach for Reputation System-Based Trust-Aware Routingin Wireless Sensor Networks,” IET Communications, vol. 3,no. 5, pp. 846–858, May 2009.

[27] A. Jøsang, “A Logic for Uncertain Probabilities,” Int. J. of Un-certainty, Fuzziness and Knowledge-Based Systems, vol. 9, no. 3,pp. 279–311, Jun 2001.

[28] A. Jøsang, “The Consensus Operator for Combining Beliefs,”Artificial Intelligence, vol. 141, no. 1, pp. 157–170, Oct 2002.

[29] Y. Sun, H. Luo, and S. K. Das, “A Trust-Based Frameworkfor Fault-Tolerant Data Aggregation in Wireless MultimediaSensor Networks,” IEEE Trans. on Dependable and Secure Com-puting, vol. 9, no. 6, pp. 785–797, Nov 2012.

[30] H. T. Friis, “A Note on a Simple Transmission Formula,” Proc.of the IRE, vol. 34, no. 5, pp. 254–256, May 1946.

[31] S. Misra, G. Mali, and A. Mondal, “Distributed Topology Man-agement for Wireless Multimedia Sensor Networks: ExploitingConnectivity and Cooperation,” Int. J. of Communication Sys-tems, vol. 28, no. 7, pp. 1367–1386, May 2015.

[32] F. Etezadi, K. Zarifi, A. Ghrayeb, and S. Affes, “DecentralizedRelay Selection Schemes in Uniformly Distributed WirelessSensor Networks,” IEEE Trans. on Wireless Communications,vol. 11, no. 3, pp. 938–951, Mar 2012.

[33] Z. Han, D. Niyato, W. Saad, T. Basar, and A. Hjorungnes, GameTheory in Wireless and Communication Networks: Theory, Models,

and Applications, 1st ed. New York, NY, USA: CambridgeUniversity Press, 2012.

[34] A. Mondal, S. Misra, and M. S. Obaidat, “Distributed HomeEnergy Management System With Storage in Smart Grid UsingGame Theory,” IEEE Systems Journal, vol. 11, no. 3, pp. 1857–1866, September 2017.

[35] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan,“Energy-Efficient Communication Protocol for Wireless Mi-crosensor Networks,” in Proc. of the 33rd Annual Hawaii Int.Conf. on System Sciences, Jan 2000, pp. 1–10.

Aishwariya Chakraborty (S’17) is presentlya pursuing her M.S. degree from Dept. ofComputer Science and Engineering, IndianInstitute of Technology Kharagpur, India. Hercurrent research interests include algorithmdesign for sensor-cloud, service-oriented ar-chitecture, and wireless sensor networks.She received her B.Tech. degree from WestBengal University of Technology in 2015. Sheis a student member of IEEE.

Ayan Mondal (S’13) is presently a TCS Fel-low and pursuing his Ph.D. degree from Dept.of Computer Science and Engineering, In-dian Institute of Technology Kharagpur, India.His current research interests include algo-rithm design for big data networks, smart gridand wireless sensor networks. He receivedhis M.S. and B.Tech. degree from IndianInstitute of Technology Kharagpur in 2014and West Bengal University of Technology in2012, respectively. He is a student member

of IEEE and ACM.

Arijit Roy (S’13) is presently a CSIR Fel-low and pursuing his Ph.D. degree from Ad-vanced Technology Development Centre, In-dian Institute of Technology Kharagpur, India.His research interests include mobile ad-hocnetworks and wireless sensor networks. Hereceived his M.S. and B.Tech. degree fromIndian Institute of Technology Kharagpur andWest Bengal University of Technology, re-spectively. He is a student member of IEEEand ACM.

Sudip Misra (SM’11) is a Professor at theIndian Institute of Technology Kharagpur. Hereceived his Ph.D. degree from Carleton Uni-versity, Ottawa, Canada. Dr. Misra is the au-thor of over 290 scholarly research papers.He has won several national and interna-tional awards including the IEEE ComSocAsia Pacific Young Researcher Award duringIEEE GLOBECOM 2012. Dr. Misra was alsoinvited to deliver keynote/invited lectures inover 30 international conferences in USA,

Canada, Europe, Asia and Africa.